The GenAI Revolution in Fintech

Anand Khetan
Anish Patil
No items found.

Anand:

Hi, everyone. Welcome to Matrix Moments. This is Aakash, Anish and Anand. In today’s podcast we’ll brainstorm about Generative AI use cases in Fintech and financial services. Why we’re doing this, there’s lot of buzz around Gen AI and we keep talking about Gen AI use cases in media, social, gaming, Ed, Health. As fin enthusiasts we were feeling a bit of FOMO and thought we should start a discussion on Gen AI use cases in fintech FS. So, Aakash, before we dive in you have been talking about Gen AI since last six months, I think that's the only thing you talk about these days. There is enough content out there on Gen AI but can you summarize for our listeners what is exciting about this tech, what has changed in last one year?

Aakash:

Surely not the trough of disillusionment but yeah, I think with AI we’re probably in the fifth or sixth hype cycles. I would say the first AI hype cycle dates as back as post World War 2. So, it has almost been the big audacious call of everyone in computer science to think of an artificial generative intelligence. So when as a fifth, sixth hype cycle and the most recent one was the deep learning hype cycle earlier in this last decade. What is surprising is that this time around if you think about where did we have this moment of hype it’s somewhere around November 2022 last year when Chat GPT was launched. And more importantly post that if you follow the pace of activity and the speed at which things have been evolving that’s just extremely phenomenal.

Going back to your question wherein the hype cycle I think so many founders asked this question saying that is it the future or is it here and now. And I think that's the main question to answer. I would say that three things which hold true. Firstly, it’s super real, it’s here and now, the utility is clear and there is no question about how you could apply it and it is evolving super-fast. Now typically if you think of any other hype cycles you do not have these answers when we say peak of expectations. Now if you think about Gen AI, I would say take a step back, all the infatuation of late has been with Gen AI, Generative AI. When I say we’re not in like a peak expectation hype cycle it’s for AI. Gen AI is just the tip of the iceberg, right, we’ve reached a point where modality transfer which is think of it as text to video, video to audio, just between those formats that is evolving super-fast and it’s going to outpace any expectations that we have, it has already outpaced all expectations.

What is happening now is the models have the ability to capture meaning, context of language, text, and which allows dealing with software to become more intelligent. So that's the first thing. Second is AI is now getting to a point wherein it has action-oriented reasoning and even this is improving at breakneck pace. There is an adage where I think some few years back Bill Gates said that humans tend to overestimate the short term and underestimate the long term. And I think that is true, so I think that is the part of the hype cycle which is true. But when it comes to AI I think in 2022 we reached a point where foundation models have truly come of age. Are we still overestimating the short term, maybe yes. Are we underestimating the long term I worry about that a lot.

And the biggest point is the duration of those short and long terms. My guess is the real applications across all sectors, all domains of and we’ll talk about it later in terms of deconstructing what it means and then becoming real and giving a 10x boost to what we can do maybe it’s a two year horizon which is when I would say it’s not truly a hype cycle. Last piece is the why now, now when anything comes out and we’ve seen this earlier with computers, with internet, smart mobiles the true inflection point is when you reach planet scale democratization. Now I-phone was not the first smart phone, there were smart phones before that but I-phone was the one which set the stage up for planet scale smart phone democratization. I think that's what’s happening today, that is the piece that I'm most excited about. It is super exciting because if you think about what it is doing. Earlier -- AI is not new, companies like Facebook,have been using AI, they’re truly AI made up companies. But to work on those deep learning machine learning, training those models it was a huge cost investment. And those companies could afford it because their yield against that technology investment was far skewed towards multiple of dollars. What has happened with Open AI and Chat GPT and all this stuff that is happening on open source it is getting democratized. Every business can now reap the benefits of that, it’s almost like you’ve taken all the training cost, you’ve taken all those costs and CapEx away effectively turn it into OpEx. Businesses can access Open AI APIs and it’s pay as you go. Open source models are becoming almost as good as closed source models and again those can be used by businesses directly and probably in fintech you will see more and more of that because the world of fin would always want to ensure that they’re not working with something where data security is at risk and they will always have the perception gap. We saw that even with the switch to on-prem-to-cloud, it was always a resistance and probably very valid. If you look at it from this lens I would say it’s here and now and the hype is very real.

Anish:

And to add to your point I think the core difference is lot of the tooling and plumbing has been happening and has been happening in the backend for quite some time now. When you give the application layer and use cases into people’s hands literally you give them a web page and say hey, type your question here and you get an answer. That is when things start to move, or adoption starts to increase because you're then telling them there is a utility and potentially future line of sight to commercial benefits and I think that's the point.

Anand:

So as a fin native person, right, and some of the viewers who are new to AI might also connect, what stood out for me as I was learning about AI is the how real and how natural the output is versus say previous technologies. And second thing is that this time around it generates new content versus see earlier versions of AI/ML where it was more preprogrammed rule based. So those were the two things which stood out for me. Moving on, Aakash, Anish, as you guys are thinking about what would be the top areas that would be disrupted by this tech, right, what framework do you follow?

Aakash:

I think personally I always like to follow, you know, just taking a step back and looking at what are the real pieces of technology which can become points of leverage. That's broadly with any piece of technology that’s how we need to think of it. I think before the intuition that most people had with AI and I think Sam also speaks about it in one of his talks, intuitively people thought that AI is first going to go after typical process automation which is more industry, low skill jobs, then potentially coding, last would be creative. If you look at it, it’s playing out exactly the opposite. So, I think intuition in these things has typically not been the right answer and that's why we can do as much crystal ball gazing as we can but if you were to take a step back and think first principles what changes. Again, multi modal transfer, text, word, image, audio, video, 3-D, whatever you want to think about it, now this is super critical because that's how if you think about any industry that’s the interaction framework, that's how people interact with anything. You mentally start with language in your head, even if you're a coder, and then you turn it into code, your thought process is still language. Which brings me to the second piece, the biggest inflection in AI today beyond the generative capabilities has been its ability to interpret and understand language. It is almost at a human scale now; some people even say that with GPT4 it could even move way beyond that. And the third piece once you put both of these together there is so much of unstructured data whether you look the fin, commerce, content, anything. There is so much amount of data that we generate we are all in a data world, knowledge world, right, but it’s all unstructured. And the hard task had always used to be that how do you bring structure to it so that it becomes useable. The moment you have powerful models the way we have now that unstructured data becomes a goldmine. So if you put all of these together I would say the key theme then becomes any activity or a job which had to be done in a business or any human to human interaction, human to business interaction which of those can move towards a point which they get automated. So I think automation is the key theme which everyone needs to think of as a framework. I would reckon most jobs would at least have some x percentage of machine automation and 100-x of human intervention, some would move to a point where it’s 100 percent machine, some would always continue being human plus machine. And which is broadly a theme of what we call a copilot. We’ve seen copilots, people in the world of coding are already familiar with copilots. I would reckon that most businesses, most economic activity would start moving to a point where you have this copilot like approach. And depending on the degree of output or the accuracy needed simple example would be if Anish has to write something about a topic which is more factual and data rich he won't rely on AI a 100 percent. But if he has to write a very simple two-line coffee AI will take up 100 percent. That's broadly the framework I use that you can augment and improve any and every process and I would rather now flip it to you guys and ask you that if you were to think of that deconstruction of impact of AI how do you think it impacts the world of financial service in fintech?

Anish:

So if you look at traditionally financial services or fintech what comes to the top of your mind? Trust, customer awareness, risk, risk return, it’s all risk return curve, compliance, these are some of the broad topics that fintech’s or financial services institutions or customers worry about. So you will see varying levels of impact of AI on all of these facets, in some it will be higher than the others or rather in some it will be incremental and some it will be truly transformational. And if you abstract it out ultimately all of financial service is movement of money. You're lending, you're saving, you're investing, you're insuring, broadly these are the activities you would do in financial services. Now if you plot all of these transactions on a 2*2, I'm a consultant so I will --

Aakash:

Yeah, we all love 2*2.

Anish:

-- force fit a matrix here. On the x axis you have the volume and the y axis you have the value and there are varying clearly four chordants and there are varying ways you can look at it. So like high volume high value transactions, B2B transactions, the IMPSs, the RTGS, the NEFT, those kind of high transactions which happen on a daily or weekly basis. There you really are focusing on reconciliation, risk, fraud, you're looking at financial planning for companies who have sort of a high velocity environment where they’re doing lot of business. So I think there there is a different axis, if you look at the low value but high volume which is what B2C payments is, right, we do like lot of transactions every day. Risk scoring, fraud, transaction classification, you know, a lot of the – for a long period it was very hard to find out in your bank statement what is the narration of a transaction, how do you classify that. Even now if you look at lot of the bank apps they ask you, we have classified it as this but is it this. Few years down the line they should just know with your location with the kind of transaction you’ve done with the merchant code lot of this would just come together and give you a weekly or a monthly P&L as to where have you sort of spent and done these transactions.

Where I think it will be really truly non incremental is the high value and low volume which is essentially your wealth, insurance, you don’t wake up every day buying an insurance policy but when you do it it’s a pretty consultative involved say, similarly with wealth. You’ll have a SIP or you’ll have a one-time investment that you think of every month and there I think traditionally the problem has been at least in India the AUM CAC equation or the GWP CAC equation is never or just started to make sense but it’s taken a lot of time to make sense. You're putting people behind the problem, you’ve a very large funnel and then you get a very small number of people who buy it, so the cost economics don’t play out. And I think here AI can play a big part.

Aakash:

Yeah, a huge role. It would actually end up playing a huge role because again where robot advisers and all have not been able to replace that human aspect. I think what changes to your point is the biggest piece of the interpretation, context, and the generative aspect of AI. You can pretty much now move those things to pure tech. Till now it used to be more of a tofu layer, you can actually move this interaction up and it’s highly possible that tomorrow you can just talk to a piece of software which is able to design your investment folio for you.

Anish:

Or at least enabled by that. So you may still have a human involved but AI has now come to a situation where reasoning, context, example, life situation, all these things can be considered before just throwing out solutions. A chat or a tofu product generally in wealth or insurance asks you a lot of questions, you don’t have to answer so many questions. There could be pre-fetch data, you already have some information about me so you’ve classified me into a cohort. I just need you can give me very contextual products by asking me very minimal questions and then focusing on the experience and then actually the right product which can actually both improve the efficiency of the sale as well as make you that you're not feeling like you're talking to a bot where just automated and it’s going to ask you the same questions. So, yeah, I think AI plus AI enabled relationship managers or AI enabled agents of these are largely agent businesses, right, they can significantly bring down the cost and sort of make these funnels as well as experiences 10x.

Anand:

Yeah. The second way to sort of segment this is if you take a value chain approach. The way I see it you can bucket it into four categories, either it’s distribution, it’s manufacturing, it’s services or it’s related infra. In each of these buckets you can think about hundreds of intake fintech AI use cases. If you think about distribution you can further break it into information inflow and outflow. Largely whatever you guys said, right, its either about taking information input from the customer, understanding it, breaking it and then giving a output in terms of product or service, here AI plays a huge role. With NLP you can understand better what the user is doing at a lower CAC and you can churn out better products. I'm already seeing some live examples with Active.AI, Active.AI doing a great job in customer engagement. There is upstocks interestingly I was seeing I can buy IPO shares through their Whatsapp bot which interacts very seamlessly and some of our well tech companies are creating some use cases around this. So that's on distribution.

On manufacturing you can break it into risk underwriting, it could be loan risk, insurance risk and then there is wealth products where you're creating portfolio. When you think about underwriting with every advent of new tech it has got in better. We’ve been talking about data as the new OIL and there is so much on structured data but we only saw a fraction of that play out when it comes to really using that to underwrite risk. I see that changing now with a real capability to break this data faster. There are companies like Yembo in the US which uses AI technology to conduct virtual surveys of insurance assets which was very interesting. In wealth you have portfolio rebalancing, algo trading.

Then when you come to services --

Anish:

Just to add to that point I think for a long time in India there used to be alternative data underwriting and robot advisory. To be very honest they’ve not delivered to their potential over the last.

Aakash:

It’s just been an unfulfilled promise.

Anish:

Yes. And ultimately underwriting in India is still CIBIL or largely thin file customer it’s hard to underwrite, thick file customer you give them a loan everybody is fighting after the same thick file customer. So underwriting definitely transformative and even the same in wealth. It’s very hard to today prove or maybe today it’s much better but a few years ago that robot advisory models were generating better returns than humans or the index. I think that we can see that going forward that there will definitely be some change.

Aakash:

I think the biggest piece is it’s not just about better returns, even if you think of the wealth and extend that most people have a different relationship with money. Everyone thinks of wealth also very differently and which is why the unfulfilled promise because how it works for Anand versus me versus you is going to be very different, the risk appetite is very different, aversion is very different. The selection of products, perception is very different and which is exactly what a wealth manager does very well today.

Anish:

He understands your mood, he knows when the visit you.

Aakash:

I think AI can actually get to that point where it can start behaving and augmenting the relationship managed very well. So I don’t see it’s going to be, it’s not just a binary outcome thing.

Anish:

Fair. It’s a personal agent basically, it knows you.

Anand:

Yeah. Third part is servicing. So any FS product when you're selling or after sale because it involves money people have lot of questions. I’ve seen this in some of our portfolio companies, you have to build large customer service teams even if your product is totally digital. Banks have it, new fintechs are facing this issue. So what you can do now is that you can achieve a lot more with this team given you can process a lot more request using Gen AI, so that's where servicing is playing out. If you look at value add services in B2C there could be pricing discovery, in B2B it could be financial forecasting, better accounting. So there is endless number of use cases when I look at servicing and the last bit is infra which is identity verification, fraud gateways, regulation compliance stack. Adyen which is a payment company in the US is doing an amazing job, they have a shopper DNA feature wherein they identify all possible fraudulent behavior a shopper can undergo using Gen AI.

Aakash:

So Anish or Anand, I think when I hear you guys when you’re talking about this it is version of the same thing that if you think of financial services lots of information, lots of data, complex machineries, complex businesses. Servicing across that value chain everywhere there is augmentation. But if you had to make it slightly more real for everyone can you think of slightly some sharp examples of where you think it is actually risk playing out to its fullest promise.

Anish:

Yeah. Maybe I can share a few examples of few startups who are doing it. I think at the very macro level you have Bloomberg GPT, you’ll realize that you could have trained it on the largest set of data but finance is very particular. So there has to be a lot of financial services context, the same term may mean different outside of financial services versus in. So I think the whole fact that it’s a LLM completely trained on finance it’s basically your Google equivalent and then what Bloomberg GPT has done is truly commendable and I think that will keep evolving, there will be a bunch of people who’ll try to do that for insurance. There is one company which we were discussing about where – you know, ultimately I want to answer for example when I'm buying a policy can I get a 5000 rupee bed cover.

Aakash:

In this place.

Anish:

In the hospital next door. I just want to type this or give a voice command that Manipal Hospital on HAL Road mein, mujhe 5000 din ka cover karega ke nahe karega. I don’t want to read 30 pages of terms and conditions, policy, exclusion, inclusion. And we can reach to a situation where it gives me a simple answer, yes or no, and if no what is the right policy for me to pick. So that's one example, like everything in insurance is so embedded in legal terminologies and long documents I don’t want to go through all of that. So that is one example. Globally if you look at use cases there is AIDA funded by Mastercard. It uses AI to help banks find out what are the revenue and cost levers and insurance companies to figure out what are their value unlocks. There is portfolio pilot which we were also using and sort of trying the other day, it’s like your individual investment coach powered by AI similar to what you were sort of saying. And they call it as bringing hedge fund tools to mass retail investors, sort of portfolio monitoring, when to get into a stock, when do you rebalance, sometimes auto rebalancing, all of that, that's an interesting use case. And even Forward Lane and to cash to our point on human led AI improving the entire sales process and value creation. So Forward Lane is actually doing AI software solutions which are helping B2B2C advisors advise their clients better. So you're integrated into a CRM, you know the clients’ situation, you know when is his red alert, when do you need to give them a next best action nudge, when is their portfolio not doing well wherein you sort of need to intervene. So it’s enabling the human do their job a little better and that’ll increase span. Earlier you were probably serving 20 clients you can serve maybe 60 clients now.

Aakash:

And the human piece is I think where more aptly applicable to financial services because we can never move to a point where you just rely on the machine. There is a regulatory concern around it, it’s so much you always want someone to be fact checking, you need that last mile to be a human touch especially let’s say wealth.

Anish:

Where you give money to the other institutions?

Aakash:

I think which is where I'm super excited about how this plays out in financial services. Like in financial services it’s a purest form of human augmentation with AI playing out.

Anand:

And that’s probably one of the reasons why in India branch banking still works, there is this ad, app se zyada aap important ho.

Anand:

But when you think who will benefit the most out of this technology do you think that people who already have a lot of data say incumbents are at a significant advantage?

Anish:

You want to take this, Kash, this is the classic value division question.

Aakash:

Classic question and classical answer.

Anish:

Depends.

[Laughter]

Aakash:

Typically the same old answer, right, that can upstarts figure out distribution faster and now you have to apply to data. I think unlike those typical innovation dilemmas over here data has an advantage and it has a slightly disproportionate advantage. So I think, yes, incumbents will derive a lot of value out of it if they approach it fast and they do are nimble about it. But I think upstarts is not that you don’t have an option, you don’t have a right to play, it’s not that. You still can find pockets where you can deliver disproportionate value even though you might be building your data corpus over time you might be starting from a zero, it might be a cold start but let’s say the insurance example. I wouldn’t be surprised if an upstart tomorrow comes and solves for it, right, because those policies and documents already exist but someone has to do the scaffolding on top, someone has to use that data, turn it into a user experience which allows to answer Anish’s question, then makes it actionable, ties the loop. I wouldn’t be surprised if an upstart does it, so I think it’s both ways. But, yeah, data wise incumbents will have an advantage.

Anand:

And a lot of problems that we discussed about there is a horizontal SaaS solution does make sense. In some places it doesn’t make sense for each and every incumbent to go out and solve the problem on their own whereas here a lot of company creation coming in horizontal Gen AI SaaS solutions.

Anish:

So, cash, talking of business, what do you advise, so what is your advice to early stage startups or growth stage startups which have a very strong. So there’s early who can act early and then there is growth who have a very solid business model and how do they think about do you implement AI, keep going continue what you're doing or is there a transformational shift you need to change or rewire your data architecture, is it a significant uprate, how do you think those choices?

Aakash:

And which is also linked to the previous question. So I think upstarts have the advantage they’re starting fresh, so the early stage folks I would say think hard about the how we deconstructed what points of leverage AI gives you and then work customer backwards saying that what are those places where I could use those points of leverage to create better distribution. You can do it across the value chain, most importantly create a better consumer experience and user experience and build capital efficiency in my own business of how I run it. Because you're starting fresh you can pretty much build it ground up. You might have a handicap on data but I don’t think it’s going to be a huge handicap, you at least have a chance to play. For incumbents or growth stage now I keep saying that even the upstarts of last seven years are effectively incumbents when you think of something as transformational as this. I think for them it is more about can they quickly move to be in AI native. And it would require some amount of rethinking, it would require restacking because when you think of data you might have all this unstructured data. The systems have been designed in a very different way, if you think fundamentally of how a large model works it works very differently from how a relational database works. It works very differently from how typical object oriented programming language works. It’s very objective basis almost black box. You will have to figure out how do you tame this beast, right, and how do you extract value out of it. Even all the data that you have even if you have the advantage you’ll have to solve for how does it get annotated, how does it get labeled, how do you actually make it work with the large models. So for incumbents I would say the most important thing here and now is get curious, get excited, start shaping your org to pursue this. And start getting hands on as soon as possible because again the typical race which is there. But I think this time the race is less about distribution it is more about who can actually get the advantage from these last moments.

Anand:

Yeah, and what we always talk about internally as well that you should designate a AI champion in your team and make sure that it is on the organization top three priorities for next couple of quarters.

Anish:

How do we think about investments in AI?

Aakash:

AI will always – we have three VCs talking about this topic and that will typically have a lot of fear of missing out not just from I think investors but even businesses thinking of where to invest, what to do. There will be a lot of hype, there will be a lot of fear, lot of greed. There’s a lot of noisy stuff which will get funded as well, I think the core principle of how at least we would think about it would still be those basics saying that what you were saying how can it be a tool for me to get to better NPS, better consumer experience. How can it actually be a revenue driver. When we’re talking about the jobs being augmented part automation, part human I don’t think of it as part automation part human just as a cost reducer I also think of that augmentation leading to actually a multiplier on the output. Most importantly I would say being very particular about how do you govern and what I said tame this beast. All of us know AIs can be hallucinating, they can make errors. Multiple pieces around how even for the same prompt or input you could get very different answers. All of that architecture and work needs to be done, guardrails need to be put, people will have to design those systems. And I think companies who are going to go after it with that approach of saying all these 4-5 things think the value chain work on that Anish’s 2*2 matrix saying where do I go heavy on saying that I want to be more AI deep, where do I think of it as a slight augmentation, where do I think of it as a very thin layer. I think founders who’ll make those judgments are the ones who’ll end up creating large value companies.

Anand:

We discussed so many use cases, if we had to pick one each which use cases you're most excited about, Anish?

Anish:

So I think wealth and portfolio management would definitely be and we’ve spoken about it. But I think for a long time we’ve never had a overall asset liability view of a human’s net worth and some companies in India have tried it to pull data from different places, you have to show it. And all of us keep discussing how we maintain the Excels and have to manually update them every month or every quarter. That is very powerful and some part of it will be solved by AA because the data pools themselves have to open up but if you can use AI to sort of do multiple things like tracking your net worth, knowing what has moved to what are the two things that I need to spend time on. I don’t want to spend time on what are the ten investments I’ve made. What are the two things that I need to action this month, this week, today. So that would be a very interesting thing for me to watch out and we keep trying out different wealth products and this will truly be transformational if we can sort of get a product which kind of not just does selection, helps you in selection and monitoring but also helps you in rebalancing and figuring through your journey, it’s also a special journey.

Aakash:

Throughout the journey.

Anish:

It’s also a temporal phenomena, it has to tell you Anand at 30, it’s a very different answer from Anand at 40. And probably you're training it with – not probably, you will be training it with your life information and it has to give you much smarter outcomes than you would have done on your own.

Aakash:

I think wealth is a very clear one, one which I'm super excited about is insurance. Now whether it be new underwriting models why should the underwriting let’s say all aspects of insurance like all the insurance why shouldn’t the underwriting be more intelligent. Even if you think of health insurance I think Rahul Chaudhary from our team who coveres health, has spoken about medical coding and how you think of claim settlement, claim rejection, all of that. Across any insurance domain you’ll start seeing the impact of those. I’ll take an example of auto again, right, you have OBDs, you have so much connected devices you're moving electric, we’re just going even more connected and data rich. Why wouldn’t claims settlement change, why wouldn’t underwriting completely transform. And distribution we anyway spoke about, right, I think that's the one I'm actually on the lookout for, someone who comes and solves the problem saying what is the right health insurance policy that I should buy.

Anand:

Yeah, in three questions.

Aakash:

In three questions.

Anish:

And by the way the image part of it, what Cash was saying important thing is image and fintech I think one of the big use cases will be just claims, the biggest problem with self-declaration of claims or putting your own photographs is you don’t know if it’s fraud or true. And if you train it over a longer period of time over multiple insurance companies or multiple types of segments, customers, I don’t see any reason why you need somebody to go and verify.

Anand:

If I had to pick one use case which I see having an immediate impact it would be debt collection. Personal loan has been exploding and somehow collections it’s in the great vertical where it is core as well as something which I see players can outsource. And there is an immediate benefit of the technology as well where you can have different elements of technology attacking DPD 0, 30, 90 so that's the most evident use case for me. But having said that we’re very early in figuring out in picking use cases.

Anish:

We forgot to give the disclaimer, Cash, the word, we’re super early and we don’t know how it will evolve.

Aakash:

Yeah. No one knows how it will evolve, you have people at one end saying that stop the work, it is going to evolve into a sentient being.

Anish:

Smarter guys are actually saying you still need to do what you need to do which is your point which is you need to select the right segment, you need to have the right GTM, you need to have the right product. All of that needs to come into place. So no discussion in fintech is complete without --

Aakash:

Risks and disclaimers.

Anish:

Yes. So one disclaimer we’ve already given. So what are the risks and contra of why AI may limit what happens in fintech versus some of the other unregulated sectors. And even if it manages to sort of penetrate what are things that founders and public at large should look out for?

Anand:

Yeah. Number one would be what Cash mentioned earlier, FS requires a lot of things to be 100 percent accurate. AI, Gen AI as it stands today, I won't classify it as 100 percent accurate in a lot of places. I can't ask Chat GPT to do my taxes. But having said that using it for copilot alongside a human you can still leverage it to save a lot of cost and add value. Second is Gen AI is a horizontal tech after all. Every horizontal tech depending upon how you use it would have positive and negatives. Can you write a better phishing mail using Gen AI the answer would be yes. That risk also gives rise to opportunities, there is more need for fraud, security, monitoring companies. So those would be the top two for me.

Anish:

I also think data privacy and regulations and we’ve still not seen generally regulations come with a lag to technology. How do you treat personally identifiable information, can you use information to be trained on models with or without customer consent. Where are you storing this data, you know in India data localization guidelines has been a big point of contention for fintechs. So I think data privacy and how you share some of this and whether open source is truly open source. You made that important point, right, like there is this debate between closed and open but how open can it be. So can my data sit in US and that data be used for something else, I don’t know. So we don’t know the limits, the problem is we don’t know the limits of what can go wrong.

Aakash:

And I think which is going to be the trickiest part. So more than risk and disclaimer I think something I believe we’ll have to live with regulation typically lacks catching up to technology innovation. Here we’re talking about something which is evolving way faster than anything we’ve seen in the past.

Anand:

Almost daily.

Aakash:

Almost daily. I think that's the one piece which makes one take a pause and maybe industry stakeholders startups and regulator all will have to make a conscious choice and say that yes, we do need to regulate but we can't stop. We can't put a halt and then decide to regulate, how can we start with like a MVP framework for regulation. It almost I think regulations will also now need to start behaving like a startup saying that this is the minimum viable, then keep going, keep going, keep going.

Anish:

Sandbox types. We already have that thought process. I think --

Aakash:

Sandbox is going to be very much needed.

Anand:

And at some places actually it would be a positive for risk and compliance. Audit trail of AI based decision making would be better than even a human. I cannot break down my old decisions and document it easily but if I'm using the AI model I can do that really well.

Anish:

Yeah. So any last thoughts?

Aakash:

No, I think you asked that, right, different stages of the venture life states or incumbents I think that is the main thought, mostly we’ll just say whether you're a founder or an operator or someone running as a say sort of big incumbent get started. Start thinking that how do you use this and going back to the earlier point, right, it is real, and it is a 10x point of leverage, it is up to you how you use it. Are there risks, yes, is it evolving too fast, yes. But the only way to keep up is going to be to start and not to wait.

Anish:

Anything from you, Anand?

Anand:

No. Just look forward to engage if you're building anything in fintech AI I would love to learn together and see if we can build something big out of it.

Anish:

Yeah. Not just build, if you're an existing startup or a CXO who’s trying to solve something in your vertical using AI and we’re always happy to brainstorm and sort of connect you to some of our other portfolio companies as well who are having some things, exciting things, in the hopper. So hit either Cash, me or Anand up and we’re happy to brainstorm. No time is too early for entrepreneurs.

Aakash:

Yeah. Thank you.

Anand:

Thank you.

Related Content

How Unsecured Lending Became More Viable in India
How Unsecured Lending Became More Viable in India
Vikram Vaidyanathan
Anish Patil
The Middle Journey of Fintech
The Middle Journey of Fintech
Vikram Vaidyanathan
Anand Khetan
Top 4 Insights from Sumit Chhazed on Disrupting the Two-Wheeler Market in India
Top 4 Insights from Sumit Chhazed on Disrupting the Two-Wheeler Market in India
Rajinder Balaraman
Anish Patil
Anand Khetan
Anish Patil
Anand Khetan
Anish Patil